Developing Patient-level Risk Prediction Models for Prescription Opioid Overdose Summary / Abstract Morbidity and mortality related to prescription opioid use and abuse are major clinical and public health problems. Reducing prescription opioid overdose rates requires efforts on multiple fronts aimed at reducing both patients? transition to long-term opioid use and their subsequent overdose risk. Prescription drug monitoring programs (PDMPs)?statewide electronic databases containing all controlled substance prescriptions and that clinicians can query in real time?are one promising tool for promoting safe opioid prescribing, but their full potential remains untapped. One reason for this is that most prior research has focused on patients? mean opioid dose, and has used either aggregate data or data restricted to specific health systems or insurers. Evidence derived from large, population-based, patient-level longitudinal data is needed to better inform national and state efforts to reduce prescription opioid-related harms. For example, high-dose opioid use is associated with greater overdose risk, but our preliminary data indicate that rate of opioid dose escalation is also an important and under-studied predictor of overdose risk. This proposal?s long-term goal is to lay the groundwork for multivariable PDMP-based risk prediction tools that clinicians and public health officials can use to assess overdose risk in the same way that Framingham-type tools are currently used to assess cardiac risk. The proposal?s overarching hypotheses are that rate of opioid dose escalation will be associated with both transition to long-term opioid use and incident opioid overdose, and that overall overdose risk will be concentrated in a relatively small group of high-risk patients. The objective of this proposal is to identify longitudinal opioid prescribing patterns associated with a) new opioid users? transition to long-term use (i.e., continual opioid use for >90 days), b) patients? incident fatal or nonfatal opioid-related overdose (including heroin overdose), and c) repeat overdose by analyzing longitudinal, patient-level prescribing and overdose data for all of California between 2008 and 2016. We will take the novel step of linking 3 statewide longitudinal databases at the patient level: prescribing data from California?s PDMP, death certificate data, and statewide hospital discharge and emergency department data. Mixed-effects regression methods for longitudinal data will be used to analyze associations between opioid prescribing patterns and proposal outcomes. Results from these analyses will be used to develop and prospectively validate clinical risk prediction models for each outcome. This project will produce validated risk prediction models derived from population-based, patient-level longitudinal data that will be used to build clinical risk prediction tools that can eventually be incorporated into PDMPs in order to inform prescribing decisions at the point of care. Results from California will also be useful to the 48 other states that have PDMPs. This project advances a research program aimed at developing and evaluating tools that clinicians can use to make safer opioid prescribing decisions, and that researchers and policymakers can use to design and evaluate clinical, public health, and policy interventions.